amber聯動orca多線程求助

wuzhiyi

我想用amber/orca來做QMMM計算，如果將  num_threads = 1,就一切正常，但如果設成  num_threads = 16,就會在运行orca的過程中退出。

orc_job.dat中顯示的錯誤是

ORCA finished by error termination in GTOInt

Calling Command: mpirun -np 16  /opt/gridware/apps/orca/orca_4_2_0_linux_x86-64_openmpi314/orca_gtoint_mpi orc_job.int.tmp orc_job

[file orca_tools/qcmsg.cpp, line 458]:

.... aborting the run

但是我直接运行mpirun -np 16  /opt/gridware/apps/orca/orca_4_2_0_linux_x86-64_openmpi314/orca_gtoint_mpi orc_job.int.tmp orc_job則一切正常
而且amber单独运行一切正常，orca单独运行也一切正常。
附上amber输入文件
Minimize waters
&cntrl
  imin = 1,      ! Perform an energy minimization
  ntx = 1,       ! Read Coordinates, but no velocities
  irest = 0,     ! Do not restart the simulation
  ntmin = 1,     ! For NCYC cycles the steepest descent method is used then conjugate gradient is switched on
  nmropt = 0,    ! No nmr-type analysis will be done
  maxcyc = 2000, ! The maximum number of cycles of minimization.
  ncyc = 1000,   ! The method of minimization will be switched from steepest descent to conjugate gradient after NCYC cycles.
  ntpr = 100,    ! Frequency of printing of the energy and its components.
  ntwr = 250,    ! Every ntwr steps during dynamics, the “restrt” file will be written.
  cut = 10.0,    ! Non-bond cut off
  ifqnt = 1,     ! Use external QM
/
&qmmm
  qmmask=':12,127,158',
  qmcharge=0,
  spin = 1,
  qm_ewald = 0,
  qm_theory = 'EXTERN',
  qmcut = 10.0, ! just use 1 nm
  qmshake = 0,
  writepdb = 1,
/
&orc
  basis = '',
  method = 'B97-3c',
  num_threads = 1,
/

wuzhiyi

好像发现只能二选一 要么orca跑16线程 要么sander跑16线程 不能两个都有跑16线程

s345538842

wuzhiyi 发表于 2019-10-26 17:32
好像发现只能二选一 要么orca跑16线程 要么sander跑16线程 不能两个都有跑16线程

好像是存在这个问题，好像建议是sander用单线程的版本

You have to use the same MPI version that you used during compilation. There are exceptions to this rule, but in general this is required.

With OpenMPI versions that I tested, it is not possible to run both sander.MPI and Orca in parallel. It is likely that this will fail. You probably have to use the serial sander version. Typically this is not a big issue since the MM computation is much faster than ab inito or DFT QM.
http://archive.ambermd.org/201803/0026.html